# Restore runbook
1. Declare the recovery point and freeze unsafe writes.
2. Restore PostgreSQL base backup and replay WAL to the target timestamp.
3. Validate constraints, counts, checksums, and application smoke tests.
4. Restore or warm Redis; never let stale cache override durable data.
5. Restore Qdrant snapshots and verify collection, shard, and replica health.
6. Restore Chroma persistent storage and validate collection counts and sample retrieval.
7. Reopen traffic progressively and record actual RPO/RTO.